— Some of the very first models were trained on a smaller number of photos, but we knew that a high-quality dataset would be the most important point and kept the focus on it. We actually manually processed a dataset of impressive size — the task was to find photos that met the criteria for training and mark the background and the person on them. Then the neural network was trained on this base.
changing hair color, there are dozens of programs on the market for changing the entire hairstyle, why do you think yours “took off”?
— There are indeed many applications, but none of them afghanistan whatsapp number list can dye hair both automatically and qualitatively. Another important factor is that we have learned not only to segment hair perfectly, but also have done a good job with color transfer.
123
— You already talked about changing skin color, but are there any plans to further develop the by-product, for example, add the ability to edit facial features?
— We are glad that the hair color changing technology has given us some PR, but we will not focus on similar features. The only thing we will update the app so that it shows you other hair colors in real time, as we can do.
— Do you plan to add AR functions?
— We are not thinking in this direction now, we were focused on other areas. Maybe in the future, we will also work on augmented reality, but I will not guess.
— You previously talked about providing SDK to Loreal and Schwarzkopf. Have you received any offers from representatives of the beauty industry?
— We are currently discussing possible areas of cooperation with several major manufacturers in this area, and the negotiations are proceeding on a rather positive note. Perhaps there will be announcements soon, but I cannot disclose this information now for reasons of confidentiality.